[WIP] Operators and Integration with cudensitymat#2407
Closed
sacpis wants to merge 248 commits intoNVIDIA:mainfrom
Closed
[WIP] Operators and Integration with cudensitymat#2407sacpis wants to merge 248 commits intoNVIDIA:mainfrom
sacpis wants to merge 248 commits intoNVIDIA:mainfrom
Conversation
sacpis
requested review from
1tnguyen and
bettinaheim
and removed request for
1tnguyen and
bettinaheim
amccaskey
reviewed
Nov 25, 2024
runtime/cudaq/dynamics/operators.h
Outdated
| @@ -0,0 +1,495 @@ | |||
| /******************************************************************************* | |||
Collaborator
There was a problem hiding this comment.
Very excited for this feature. I appreciate all the work you two have done on this one. One thought here -
Operators are more broad that dynamics. Ideally this would be part of a cudaq/operators.h public header. Did we ever finish the language review for this feature? Would be good to get a meeting scheduled.
|
CUDA Quantum Docs Bot: A preview of the documentation can be found here. |
1tnguyen
reviewed
Jan 23, 2025
1tnguyen
reviewed
Jan 23, 2025
1tnguyen
reviewed
Jan 23, 2025
1tnguyen
reviewed
Jan 23, 2025
sacpis
changed the title
1tnguyen
reviewed
Jan 29, 2025
unittests/CMakeLists.txt
Outdated
Comment on lines
+47
to
+57
| dynamics/scalar_ops_simple.cpp | ||
| dynamics/scalar_ops_arithmetic.cpp | ||
| dynamics/elementary_ops_simple.cpp | ||
| dynamics/elementary_ops_arithmetic.cpp | ||
| dynamics/product_operators_arithmetic.cpp | ||
| dynamics/test_runge_kutta_time_stepper.cpp | ||
| dynamics/test_runge_kutta_integrator.cpp | ||
| dynamics/test_helpers.cpp | ||
| dynamics/rydberg_hamiltonian.cpp | ||
| dynamics/test_cudm_helpers.cpp | ||
| dynamics/test_cudm_state.cpp | ||
| dynamics/test_cudm_time_stepper.cpp |
Collaborator
There was a problem hiding this comment.
These dynamics tests are very different from the usual CUDAQ_RUNTIME_TEST_SOURCES so we should move them into a separate target. Adding them here will cause these tests to be repeated for all simulator backends.
1tnguyen
reviewed
Jan 29, 2025
| # Make it so we can get function symbols | ||
| set (CMAKE_ENABLE_EXPORTS TRUE) | ||
|
|
||
| include_directories(/usr/local/cuda/targets/x86_64-linux/include) |
Collaborator
There was a problem hiding this comment.
The CUDA include and lib paths should be detected using CMake. We probably want to wrap it in a CUDA_FOUND block as well since not all build environment has CUDA.
1tnguyen
reviewed
Jan 29, 2025
unittests/CMakeLists.txt
Outdated
| $ENV{CUQUANTUM_INSTALL_PREFIX}/lib/libcudensitymat.so.0 | ||
| /usr/local/cuda-12.0/targets/x86_64-linux/lib/libcudart.so.12) |
Collaborator
There was a problem hiding this comment.
After we do the refactoring in https://github.com/NVIDIA/cuda-quantum/pull/2407/files#r1933216013, we should only add these link libs to the dynamics unit test target.
1tnguyen
reviewed
Jan 29, 2025
Comment on lines
+418
to
+435
| gtest_main | ||
| $ENV{CUQUANTUM_INSTALL_PREFIX}/lib/libcudensitymat.so.0) |
Collaborator
There was a problem hiding this comment.
We shouldn't need to add this to existing unit tests.
1tnguyen
reviewed
Jan 29, 2025
| cudm_state::cudm_state(std::vector<std::complex<double>> rawData) | ||
| : rawData_(rawData), state_(nullptr), handle_(nullptr), | ||
| hilbertSpaceDims_() { | ||
| HANDLE_CUDM_ERROR(cudensitymatCreate(&handle_)); |
Collaborator
There was a problem hiding this comment.
We should only have a single cudensitymatHandle_t throughout the code (i.e., a single cudensitymatCreate and cudensitymatDestroy).
For example, we already create a handle in test_cudm_time_stepper.cpp (SetUp); thus creating another handle here is problematic.
I'd suggest passing the handle here, e.g.
cudm_state::cudm_state(cudensitymatHandle_t handle,
const std::vector<std::complex<double>> &rawData)
: rawData_(rawData), state_(nullptr), handle_(handle), hilbertSpaceDims_()
1tnguyen
reviewed
Jan 29, 2025
Comment on lines
+28
to
+36
| if (handle_) { | ||
| cudensitymatDestroy(handle_); | ||
| } |
Collaborator
There was a problem hiding this comment.
Like above, we shouldn't be creating and destroying the handle during state object life cycle.
1tnguyen
reviewed
Jan 29, 2025
Comment on lines
+169
to
+171
| rawComponentData[i] = | ||
| reinterpret_cast<std::complex<double> *>(rawData_.data()) + offset; | ||
| componentBuffers[i] = static_cast<void *>(rawComponentData[i]); |
Collaborator
There was a problem hiding this comment.
We're supposed to attached device memory here (rawData_ is a host array).
I think we need to add a gpuData_ member var and perform cudaMalloc + cudaMemcpy.
1tnguyen
reviewed
Jan 29, 2025
| if (!liouvillian_) { | ||
| throw std::runtime_error("Liouvillian is not initialized."); | ||
| } | ||
|
|
Collaborator
There was a problem hiding this comment.
It looks like we're missing some steps here:
cudensitymatOperatorPrepareActioncudensitymatWorkspaceGetMemorySizecudensitymatWorkspaceSetMemory
1tnguyen
reviewed
Jan 29, 2025
|
|
||
| std::cout << "Update the state ..." << std::endl; | ||
| // Update the state | ||
| state = std::move(next_state); |
Collaborator
There was a problem hiding this comment.
Should we update the input state or return the action result (RHS) in this function?
We might need the latter since the RHS might be added to the input state with some scaling depending on the integration algorithm.
1tnguyen
reviewed
Jan 29, 2025
* add skeleton of new API functions without connecting to docs yet Signed-off-by: A.M. Santana <anthonys@nvidia.com> * updates Signed-off-by: A.M. Santana <anthonys@nvidia.com> * boilerplate Signed-off-by: A.M. Santana <anthonys@nvidia.com> * clean up linking bug Signed-off-by: A.M. Santana <anthonys@nvidia.com> * adding schedule implementation * fixing typo * small updates Signed-off-by: A.M. Santana <anthonys@nvidia.com> * Cleaning up docs preview for PR #6. * push initial tests that show memory leak in current translation to complex matrix Signed-off-by: A.M. Santana <anthonys@nvidia.com> * completely work around eigen in default elementary ops Signed-off-by: A.M. Santana <anthonys@nvidia.com> * storing changes Signed-off-by: A.M. Santana <anthonys@nvidia.com> * start to build out callback function class Signed-off-by: A.M. Santana <anthonys@nvidia.com> * working function wrapper implementation Signed-off-by: A.M. Santana <anthonys@nvidia.com> * implement complex matrix equality operator Signed-off-by: A.M. Santana <anthonys@nvidia.com> * push with broken to matrix overload Signed-off-by: A.M. Santana <anthonys@nvidia.com> * fix scoping issue found in to matrix Signed-off-by: A.M. Santana <anthonys@nvidia.com> * fill out unit tests for to matrix overload Signed-off-by: A.M. Santana <anthonys@nvidia.com> * add skeleton of new API functions without connecting to docs yet Signed-off-by: A.M. Santana <anthonys@nvidia.com> * updates Signed-off-by: A.M. Santana <anthonys@nvidia.com> * boilerplate Signed-off-by: A.M. Santana <anthonys@nvidia.com> * clean up linking bug Signed-off-by: A.M. Santana <anthonys@nvidia.com> * initial scalar value support and tests Signed-off-by: A.M. Santana <anthonys@nvidia.com> * update to_value to evaluate to amtch python api Signed-off-by: A.M. Santana <anthonys@nvidia.com> * adding kwargs capability in C++ by using std::variant and std::bind * renaming VariantArg to NumericType to match Python * adding a std::variant returntype nad adjusting the test accordingly * Cleaning up docs preview for PR #7. * commit first draft of arithmetic against complex doubles Signed-off-by: A.M. Santana <anthonys@nvidia.com> * little build errors Signed-off-by: A.M. Santana <anthonys@nvidia.com> * call generator directly instead of evaluate in operator overloads Signed-off-by: A.M. Santana <anthonys@nvidia.com> * remove old constructor Signed-off-by: A.M. Santana <anthonys@nvidia.com> * copy constructor Signed-off-by: A.M. Santana <anthonys@nvidia.com> * push partially working arithmetic operations Signed-off-by: A.M. Santana <anthonys@nvidia.com> * comment back in tests Signed-off-by: A.M. Santana <anthonys@nvidia.com> * remove constructor that takes removed parameters member Signed-off-by: A.M. Santana <anthonys@nvidia.com> * still having memory issues Signed-off-by: A.M. Santana <anthonys@nvidia.com> * implement and test remaining pre defined elementary ops except displace and squeeze. implement complex matrix exponential Signed-off-by: A.M. Santana <anthonys@nvidia.com> * fix for segfault in copy constructor but still have memory issues from arithmetic Signed-off-by: A.M. Santana <anthonys@nvidia.com> * temp patch to get scalar arithmetic against complex doubles working. against doubles wrapped in functions still broken Signed-off-by: A.M. Santana <anthonys@nvidia.com> * potential fix for callback function going out of scope Signed-off-by: A.M. Santana <anthonys@nvidia.com> * confirm fix for scalar ops from functions and reenable tests Signed-off-by: A.M. Santana <anthonys@nvidia.com> * begin to support scalar against scalar ops and simplify other arithmetic definitions with macros Signed-off-by: A.M. Santana <anthonys@nvidia.com> * support for remaining arithmetic against other scalar ops except for assignment operators Signed-off-by: A.M. Santana <anthonys@nvidia.com> * add checks to ensure local variables are picked up fine in generator functions Signed-off-by: A.M. Santana <anthonys@nvidia.com> * clean up test file Signed-off-by: A.M. Santana <anthonys@nvidia.com> * full refactor to take a [arameter map with elementary operator implementation back under construction Signed-off-by: A.M. Santana <anthonys@nvidia.com> * fix elementary ops Signed-off-by: A.M. Santana <anthonys@nvidia.com> * start to manually merge in operator sum changes Signed-off-by: A.M. Santana <anthonys@nvidia.com> * camel case some things and underscore some others Signed-off-by: A.M. Santana <anthonys@nvidia.com> * fix build errors from header file Signed-off-by: A.M. Santana <anthonys@nvidia.com> * refactor implementation file names, delete unused dynamics.h, refactor unittests with new dynamics folder Signed-off-by: A.M. Santana <anthonys@nvidia.com> * storing large set of changes to dynamics folder structure and implementing more arithmetic Signed-off-by: A.M. Santana <anthonys@nvidia.com> * add cudaq tensor type Signed-off-by: Alex McCaskey <amccaskey@nvidia.com> * add uint8 tensor to python api Signed-off-by: Alex McCaskey <amccaskey@nvidia.com> * Update unittests/utils/tensor_tester.cpp Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update unittests/utils/tensor_tester.cpp Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update unittests/utils/tensor_tester.cpp Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update runtime/cudaq/utils/details/impls/xtensor_impl.cpp Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update runtime/cudaq/utils/details/impls/xtensor_impl.cpp Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update runtime/cudaq/utils/details/impls/xtensor_impl.cpp Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update python/runtime/utils/py_tensor.cpp Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update runtime/cudaq/utils/extension_point.h Co-authored-by: Ben Howe <141149032+bmhowe23@users.noreply.github.com> Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update runtime/cudaq/utils/extension_point.h Co-authored-by: Ben Howe <141149032+bmhowe23@users.noreply.github.com> Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update runtime/cudaq/utils/tensor.h Co-authored-by: Ben Howe <141149032+bmhowe23@users.noreply.github.com> Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update runtime/cudaq/utils/tensor.h Co-authored-by: Ben Howe <141149032+bmhowe23@users.noreply.github.com> Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update runtime/cudaq/utils/details/impls/xtensor_impl.cpp Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update runtime/cudaq/utils/details/tensor_impl.h Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update runtime/cudaq/utils/details/tensor_impl.h Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update runtime/cudaq/utils/details/tensor_impl.h Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Spelling fixes. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * push current state Signed-off-by: A.M. Santana <anthonys@nvidia.com> * Format Signed-off-by: Anna Gringauze <agringauze@nvidia.com> * Format Signed-off-by: Anna Gringauze <agringauze@nvidia.com> * Drop debug code. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Cleanup tensor types a tad. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Fix spelling Signed-off-by: Anna Gringauze <agringauze@nvidia.com> * Fix spelling and format python code Signed-off-by: Anna Gringauze <agringauze@nvidia.com> * Sort spelling allowlist Signed-off-by: Anna Gringauze <agringauze@nvidia.com> * Cleaning up docs preview for PR #9. * more arithmetic support Signed-off-by: A.M. Santana <anthonys@nvidia.com> * Update spelling and formatting Signed-off-by: Anna Gringauze <agringauze@nvidia.com> * more implementation and more tests Signed-off-by: A.M. Santana <anthonys@nvidia.com> * Update spelling and formatting Signed-off-by: Anna Gringauze <agringauze@nvidia.com> * This is an attempt to sort out the different ownership models of tensor data. The python stub for "take" needs to be implemented however. The semantics are: - copy : the tensor object makes a copy of the data and owns the copy. i.e. there is a unique_ptr. - take : the tensor object steals a copy of the data from the caller. In order for this case to make sense, we want the caller to guarantee that the data is unique before we steal it. This can be done by forcing the client code to wrap the tensor data in a unique_ptr *before* the take call. - borrow : In this case, the tensor object has no ownership of the tensor data and just naively assumes the client will manage the data correctly. In this case, a raw pointer to the client's data is used. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Update the missing py_tensor code. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Add handling for empty shape case. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Add more python tests Signed-off-by: Anna Gringauze <agringauze@nvidia.com> * Add a take() with move semantics. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * fix product operator constructor issue and tests Signed-off-by: A.M. Santana <anthonys@nvidia.com> * fix more test more Signed-off-by: A.M. Santana <anthonys@nvidia.com> * cover all arithmetic Signed-off-by: A.M. Santana <anthonys@nvidia.com> * update before merging in tensor pr Signed-off-by: A.M. Santana <anthonys@nvidia.com> * Fix __init__ argument order so tests pass. The shape must come first. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Remove StateTensor. (Not used.) Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Add more comments on how to use this stuff. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Added python tests and fixed some issues * Merge with tensor Signed-off-by: Anna Gringauze <agringauze@nvidia.com> * DCO Remediation Commit for Anna Gringauze <agringauze@nvidia.com> I, Anna Gringauze <agringauze@nvidia.com>, hereby add my Signed-off-by to this commit: e78def4 Signed-off-by: Anna Gringauze <agringauze@nvidia.com> * Add some boilerplate for tensor operators. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Make the compiler work a bit harder. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Add move constructor. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Make sure the return values for operators work as expected. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * clang-format Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Add override to dtor. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Add forward decls. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * More fussy templates. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Workaround warnings from g++. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Fix typos. Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> * Support copy semantics for Numpy 2.0 * DCO Remediation Commit for Anna Gringauze <agringauze@nvidia.com> I, Anna Gringauze <agringauze@nvidia.com>, hereby add my Signed-off-by to this commit: 85fae37 Signed-off-by: Anna Gringauze <agringauze@nvidia.com> * Try fixing doc gen and c++ compilation errors Signed-off-by: Anna Gringauze <agringauze@nvidia.com> * Add compilation test for nvcc Signed-off-by: Anna Gringauze <agringauze@nvidia.com> * Remove temp file Signed-off-by: Anna Gringauze <agringauze@nvidia.com> * store changes Signed-off-by: A.M. Santana <anthonys@nvidia.com> * first pass of updating return types to cudaq tensor Signed-off-by: A.M. Santana <anthonys@nvidia.com> * store working version with tests before rebase Signed-off-by: A.M. Santana <anthonys@nvidia.com> * fix improper merge resolution issues Signed-off-by: A.M. Santana <anthonys@nvidia.com> * more issues with resolving merge Signed-off-by: A.M. Santana <anthonys@nvidia.com> * first pass of translation to matrix_2 with build errors fixed. now double checking tests Signed-off-by: A.M. Santana <anthonys@nvidia.com> * fix remaining artifacts from rebase Signed-off-by: A.M. Santana <anthonys@nvidia.com> * fix matrix checks for simple elementary op unit tests Signed-off-by: A.M. Santana <anthonys@nvidia.com> * first pass of implementing deeper matrix checks in tests. Have two files left to finish Signed-off-by: A.M. Santana <anthonys@nvidia.com> * fix copyright headers Signed-off-by: A.M. Santana <anthonys@nvidia.com> * minor change to check verified commit Signed-off-by: A.M. Santana <anthonys@nvidia.com> --------- Signed-off-by: A.M. Santana <anthonys@nvidia.com> Signed-off-by: Alex McCaskey <amccaskey@nvidia.com> Signed-off-by: Eric Schweitz <eschweitz@nvidia.com> Signed-off-by: Anna Gringauze <agringauze@nvidia.com> Co-authored-by: Sachin Pisal <spisal@nvidia.com> Co-authored-by: cuda-quantum-bot <cuda-quantum-bot@users.noreply.github.com> Co-authored-by: Alex McCaskey <amccaskey@nvidia.com> Co-authored-by: Eric Schweitz <eschweitz@nvidia.com> Co-authored-by: Ben Howe <141149032+bmhowe23@users.noreply.github.com> Co-authored-by: Anna Gringauze <agringauze@nvidia.com>
Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Thien Nguyen <thiennguyen@nvidia.com>
Signed-off-by: Thien Nguyen <thiennguyen@nvidia.com>
Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Sachin Pisal <spisal@nvidia.com>
* Adding mock function to create a matrix operator * Adding unittests for mul * Formatting Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Sachin Pisal <spisal@nvidia.com>
* Adding function to get space_mode_extents * Adding more unittests for op_conversion Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Sachin Pisal <spisal@nvidia.com>
…h a scalar Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Thien Nguyen <thiennguyen@nvidia.com>
…da-quantum into dynamics_cpp_operators
…conversion Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Thien Nguyen <thiennguyen@nvidia.com>
* Correcting tensor callback function name Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Thien Nguyen <thiennguyen@nvidia.com>
* Moved other reponsibilities like flattening the matrix, gettting subspace extents to the create elementary operator function * Unittests updated * Wrapped cudm_helper in a class Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Sachin Pisal <spisal@nvidia.com>
Signed-off-by: Thien Nguyen <thiennguyen@nvidia.com>
Signed-off-by: Sachin Pisal <spisal@nvidia.com>
…trix operator Signed-off-by: Sachin Pisal <spisal@nvidia.com>
… from elementary ops Signed-off-by: Thien Nguyen <thiennguyen@nvidia.com>
…tor_to_term signature Signed-off-by: Sachin Pisal <spisal@nvidia.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adding missing operations in operators
Co-authored with @anthony-santana.